GPU kernel optimization beyond SASS latency tables
A post published on November 25, 2025 questions the value of SASS latency tables extracted with nvdisasm and cites 3–4% gains from instruction reordering, against a theoretical limit of 10%.
Source: Tabelas de latência SASS e reordenação de instruções (redplait.blogspot.com). Text prepared with AI from this source.
What happened and what to do
Published on November 25, 2025, the post argues that SASS latency tables extracted with nvdisasm are not useful for optimizing GPU kernels. According to the post, reordering instructions can improve performance by 3–4%, with a theoretical limit of 10%. It addresses the use of latency data and instruction scheduling in kernel optimization.
A company can apply this discussion to kernels that consume significant processing time: establish reproducible benchmarks, compare versions with different schedules, and monitor performance and regressions under representative workloads. Decisions should rely on measurements on the company’s hardware and operating scenarios, not on latency tables alone.
How the consultancy can help
Wendelmaques can diagnose optimization opportunities in kernels and performance measurement, scope an implementation with benchmarks and monitoring, and support operation of the solution.
Next step
Send a short description of the kernel or GPU workload you want to optimize to receive a scoped proposal.
Let's assess your project
I reply with scope and a proposal.